minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 13 Packages Projects Releases Wiki Activity

[Test] Skip STANDALONE speculative decoding tests for different hidden sizes (#14733)

Browse Source
This commit is contained in:
Alison Shao
2025-12-09 22:21:35 +09:00
committed by GitHub
parent 98c430e114
commit 13680e5542
1 changed files with 8 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/srt/models/test_nvidia_nemotron_nano_v2.py
+8
View File
@@ -24,6 +24,10 @@ class TestNvidiaNemotronNanoV2NVFP4(GSM8KMixin, CustomTestCase):
other_args = ["--max-mamba-cache-size", "256"] other_args = ["--max-mamba-cache-size", "256"]
@unittest.skip(
"STANDALONE speculative decoding does not yet support target and draft models "
"with different hidden sizes (Nemotron-9B: 4480, Llama-3.2-1B: 2048)"
)
class TestNvidiaNemotronNanoV2SpeculativeDecoding(GSM8KMixin, CustomTestCase): class TestNvidiaNemotronNanoV2SpeculativeDecoding(GSM8KMixin, CustomTestCase):
accuracy = 0.87 accuracy = 0.87
model = "nvidia/NVIDIA-Nemotron-Nano-9B-v2" model = "nvidia/NVIDIA-Nemotron-Nano-9B-v2"
@@ -49,6 +53,10 @@ class TestNvidiaNemotronNanoV2SpeculativeDecoding(GSM8KMixin, CustomTestCase):
] ]
@unittest.skip(
"STANDALONE speculative decoding does not yet support target and draft models "
"with different hidden sizes (Nemotron-9B: 4480, Llama-3.2-1B: 2048)"
)
class TestNvidiaNemotronNanoV2SpeculativeDecodingBF16Cache(GSM8KMixin, CustomTestCase): class TestNvidiaNemotronNanoV2SpeculativeDecodingBF16Cache(GSM8KMixin, CustomTestCase):
accuracy = 0.87 accuracy = 0.87
model = "nvidia/NVIDIA-Nemotron-Nano-9B-v2" model = "nvidia/NVIDIA-Nemotron-Nano-9B-v2"